# Topic classification: is zero-shot the way?

**URL:** </static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftopic-classification-is-zero-shot-the-way%2F9180%3E%5C
**Category:** Beginners\
**Created:** [August 12, 2021, 9:35am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftopic-classification-is-zero-shot-the-way%2F9180 "2021-08-12T09:35:23Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Neuroinformatica](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/neuroinformatica/32/36578_2.png) [@Neuroinformatica](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FNeuroinformatica)%5C
**Post date:** [August 12, 2021, 9:35am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftopic-classification-is-zero-shot-the-way%2F9180%2F1 "2021-08-12T09:35:23Z")

</div>

Hello there

I need to extract info from italian medical records. Since such documents are somewhat long (2-3k words), my idea is to split them in subsections, and then try to classify the topic of each subsection. This would allow me to be more confident about info extraction (i.e. I am extracting a specific info from the correct subsection).

I was wondering if zero-shot classification is the best tool to do this. Moreover, since the content of records is very specific (exams results, medical reports and so on), I think I will need to do a fine-tuning of some kind.  
If this is correct, how could I do that? And approximately how many data I would need?

Thanks a lot